4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 #include <sys/zfs_context.h>
32 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/vdev_impl.h>
38 #include <sys/uberblock_impl.h>
39 #include <sys/metaslab.h>
40 #include <sys/metaslab_impl.h>
41 #include <sys/space_map.h>
42 #include <sys/space_reftree.h>
45 #include <sys/fs/zfs.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/trim_map.h>
51 SYSCTL_DECL(_vfs_zfs);
52 SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV");
55 * Virtual device management.
59 * The limit for ZFS to automatically increase a top-level vdev's ashift
60 * from logical ashift to physical ashift.
62 * Example: one or more 512B emulation child vdevs
63 * child->vdev_ashift = 9 (512 bytes)
64 * child->vdev_physical_ashift = 12 (4096 bytes)
65 * zfs_max_auto_ashift = 11 (2048 bytes)
66 * zfs_min_auto_ashift = 9 (512 bytes)
68 * On pool creation or the addition of a new top-level vdev, ZFS will
69 * increase the ashift of the top-level vdev to 2048 as limited by
70 * zfs_max_auto_ashift.
72 * Example: one or more 512B emulation child vdevs
73 * child->vdev_ashift = 9 (512 bytes)
74 * child->vdev_physical_ashift = 12 (4096 bytes)
75 * zfs_max_auto_ashift = 13 (8192 bytes)
76 * zfs_min_auto_ashift = 9 (512 bytes)
78 * On pool creation or the addition of a new top-level vdev, ZFS will
79 * increase the ashift of the top-level vdev to 4096 to match the
80 * max vdev_physical_ashift.
82 * Example: one or more 512B emulation child vdevs
83 * child->vdev_ashift = 9 (512 bytes)
84 * child->vdev_physical_ashift = 9 (512 bytes)
85 * zfs_max_auto_ashift = 13 (8192 bytes)
86 * zfs_min_auto_ashift = 12 (4096 bytes)
88 * On pool creation or the addition of a new top-level vdev, ZFS will
89 * increase the ashift of the top-level vdev to 4096 to match the
90 * zfs_min_auto_ashift.
92 static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT;
93 static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT;
96 sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS)
101 val = zfs_max_auto_ashift;
102 err = sysctl_handle_64(oidp, &val, 0, req);
103 if (err != 0 || req->newptr == NULL)
106 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift)
109 zfs_max_auto_ashift = val;
113 SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift,
114 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
115 sysctl_vfs_zfs_max_auto_ashift, "QU",
116 "Max ashift used when optimising for logical -> physical sectors size on "
117 "new top-level vdevs.");
120 sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS)
125 val = zfs_min_auto_ashift;
126 err = sysctl_handle_64(oidp, &val, 0, req);
127 if (err != 0 || req->newptr == NULL)
130 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift)
133 zfs_min_auto_ashift = val;
137 SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift,
138 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t),
139 sysctl_vfs_zfs_min_auto_ashift, "QU",
140 "Min ashift used when creating new top-level vdevs.");
142 static vdev_ops_t *vdev_ops_table[] = {
161 * When a vdev is added, it will be divided into approximately (but no
162 * more than) this number of metaslabs.
164 int metaslabs_per_vdev = 200;
165 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, metaslabs_per_vdev, CTLFLAG_RDTUN,
166 &metaslabs_per_vdev, 0,
167 "When a vdev is added, how many metaslabs the vdev should be divided into");
170 * Given a vdev type, return the appropriate ops vector.
173 vdev_getops(const char *type)
175 vdev_ops_t *ops, **opspp;
177 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
178 if (strcmp(ops->vdev_op_type, type) == 0)
185 * Default asize function: return the MAX of psize with the asize of
186 * all children. This is what's used by anything other than RAID-Z.
189 vdev_default_asize(vdev_t *vd, uint64_t psize)
191 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
194 for (int c = 0; c < vd->vdev_children; c++) {
195 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
196 asize = MAX(asize, csize);
203 * Get the minimum allocatable size. We define the allocatable size as
204 * the vdev's asize rounded to the nearest metaslab. This allows us to
205 * replace or attach devices which don't have the same physical size but
206 * can still satisfy the same number of allocations.
209 vdev_get_min_asize(vdev_t *vd)
211 vdev_t *pvd = vd->vdev_parent;
214 * If our parent is NULL (inactive spare or cache) or is the root,
215 * just return our own asize.
218 return (vd->vdev_asize);
221 * The top-level vdev just returns the allocatable size rounded
222 * to the nearest metaslab.
224 if (vd == vd->vdev_top)
225 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
228 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
229 * so each child must provide at least 1/Nth of its asize.
231 if (pvd->vdev_ops == &vdev_raidz_ops)
232 return ((pvd->vdev_min_asize + pvd->vdev_children - 1) /
235 return (pvd->vdev_min_asize);
239 vdev_set_min_asize(vdev_t *vd)
241 vd->vdev_min_asize = vdev_get_min_asize(vd);
243 for (int c = 0; c < vd->vdev_children; c++)
244 vdev_set_min_asize(vd->vdev_child[c]);
248 vdev_lookup_top(spa_t *spa, uint64_t vdev)
250 vdev_t *rvd = spa->spa_root_vdev;
252 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
254 if (vdev < rvd->vdev_children) {
255 ASSERT(rvd->vdev_child[vdev] != NULL);
256 return (rvd->vdev_child[vdev]);
263 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
267 if (vd->vdev_guid == guid)
270 for (int c = 0; c < vd->vdev_children; c++)
271 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
279 vdev_count_leaves_impl(vdev_t *vd)
283 if (vd->vdev_ops->vdev_op_leaf)
286 for (int c = 0; c < vd->vdev_children; c++)
287 n += vdev_count_leaves_impl(vd->vdev_child[c]);
293 vdev_count_leaves(spa_t *spa)
295 return (vdev_count_leaves_impl(spa->spa_root_vdev));
299 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
301 size_t oldsize, newsize;
302 uint64_t id = cvd->vdev_id;
304 spa_t *spa = cvd->vdev_spa;
306 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
307 ASSERT(cvd->vdev_parent == NULL);
309 cvd->vdev_parent = pvd;
314 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
316 oldsize = pvd->vdev_children * sizeof (vdev_t *);
317 pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
318 newsize = pvd->vdev_children * sizeof (vdev_t *);
320 newchild = kmem_zalloc(newsize, KM_SLEEP);
321 if (pvd->vdev_child != NULL) {
322 bcopy(pvd->vdev_child, newchild, oldsize);
323 kmem_free(pvd->vdev_child, oldsize);
326 pvd->vdev_child = newchild;
327 pvd->vdev_child[id] = cvd;
329 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
330 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
333 * Walk up all ancestors to update guid sum.
335 for (; pvd != NULL; pvd = pvd->vdev_parent)
336 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
340 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
343 uint_t id = cvd->vdev_id;
345 ASSERT(cvd->vdev_parent == pvd);
350 ASSERT(id < pvd->vdev_children);
351 ASSERT(pvd->vdev_child[id] == cvd);
353 pvd->vdev_child[id] = NULL;
354 cvd->vdev_parent = NULL;
356 for (c = 0; c < pvd->vdev_children; c++)
357 if (pvd->vdev_child[c])
360 if (c == pvd->vdev_children) {
361 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
362 pvd->vdev_child = NULL;
363 pvd->vdev_children = 0;
367 * Walk up all ancestors to update guid sum.
369 for (; pvd != NULL; pvd = pvd->vdev_parent)
370 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
374 * Remove any holes in the child array.
377 vdev_compact_children(vdev_t *pvd)
379 vdev_t **newchild, *cvd;
380 int oldc = pvd->vdev_children;
383 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
385 for (int c = newc = 0; c < oldc; c++)
386 if (pvd->vdev_child[c])
389 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
391 for (int c = newc = 0; c < oldc; c++) {
392 if ((cvd = pvd->vdev_child[c]) != NULL) {
393 newchild[newc] = cvd;
394 cvd->vdev_id = newc++;
398 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
399 pvd->vdev_child = newchild;
400 pvd->vdev_children = newc;
404 * Allocate and minimally initialize a vdev_t.
407 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
411 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
413 if (spa->spa_root_vdev == NULL) {
414 ASSERT(ops == &vdev_root_ops);
415 spa->spa_root_vdev = vd;
416 spa->spa_load_guid = spa_generate_guid(NULL);
419 if (guid == 0 && ops != &vdev_hole_ops) {
420 if (spa->spa_root_vdev == vd) {
422 * The root vdev's guid will also be the pool guid,
423 * which must be unique among all pools.
425 guid = spa_generate_guid(NULL);
428 * Any other vdev's guid must be unique within the pool.
430 guid = spa_generate_guid(spa);
432 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
437 vd->vdev_guid = guid;
438 vd->vdev_guid_sum = guid;
440 vd->vdev_state = VDEV_STATE_CLOSED;
441 vd->vdev_ishole = (ops == &vdev_hole_ops);
443 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
446 mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL);
447 for (int t = 0; t < DTL_TYPES; t++) {
448 vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
451 txg_list_create(&vd->vdev_ms_list, spa,
452 offsetof(struct metaslab, ms_txg_node));
453 txg_list_create(&vd->vdev_dtl_list, spa,
454 offsetof(struct vdev, vdev_dtl_node));
455 vd->vdev_stat.vs_timestamp = gethrtime();
463 * Allocate a new vdev. The 'alloctype' is used to control whether we are
464 * creating a new vdev or loading an existing one - the behavior is slightly
465 * different for each case.
468 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
473 uint64_t guid = 0, islog, nparity;
476 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
478 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
479 return (SET_ERROR(EINVAL));
481 if ((ops = vdev_getops(type)) == NULL)
482 return (SET_ERROR(EINVAL));
485 * If this is a load, get the vdev guid from the nvlist.
486 * Otherwise, vdev_alloc_common() will generate one for us.
488 if (alloctype == VDEV_ALLOC_LOAD) {
491 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
493 return (SET_ERROR(EINVAL));
495 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
496 return (SET_ERROR(EINVAL));
497 } else if (alloctype == VDEV_ALLOC_SPARE) {
498 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
499 return (SET_ERROR(EINVAL));
500 } else if (alloctype == VDEV_ALLOC_L2CACHE) {
501 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
502 return (SET_ERROR(EINVAL));
503 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
504 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
505 return (SET_ERROR(EINVAL));
509 * The first allocated vdev must be of type 'root'.
511 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
512 return (SET_ERROR(EINVAL));
515 * Determine whether we're a log vdev.
518 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
519 if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
520 return (SET_ERROR(ENOTSUP));
522 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
523 return (SET_ERROR(ENOTSUP));
526 * Set the nparity property for RAID-Z vdevs.
529 if (ops == &vdev_raidz_ops) {
530 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
532 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
533 return (SET_ERROR(EINVAL));
535 * Previous versions could only support 1 or 2 parity
539 spa_version(spa) < SPA_VERSION_RAIDZ2)
540 return (SET_ERROR(ENOTSUP));
542 spa_version(spa) < SPA_VERSION_RAIDZ3)
543 return (SET_ERROR(ENOTSUP));
546 * We require the parity to be specified for SPAs that
547 * support multiple parity levels.
549 if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
550 return (SET_ERROR(EINVAL));
552 * Otherwise, we default to 1 parity device for RAID-Z.
559 ASSERT(nparity != -1ULL);
561 vd = vdev_alloc_common(spa, id, guid, ops);
563 vd->vdev_islog = islog;
564 vd->vdev_nparity = nparity;
566 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
567 vd->vdev_path = spa_strdup(vd->vdev_path);
568 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
569 vd->vdev_devid = spa_strdup(vd->vdev_devid);
570 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
571 &vd->vdev_physpath) == 0)
572 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
573 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
574 vd->vdev_fru = spa_strdup(vd->vdev_fru);
577 * Set the whole_disk property. If it's not specified, leave the value
580 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
581 &vd->vdev_wholedisk) != 0)
582 vd->vdev_wholedisk = -1ULL;
585 * Look for the 'not present' flag. This will only be set if the device
586 * was not present at the time of import.
588 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
589 &vd->vdev_not_present);
592 * Get the alignment requirement.
594 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
597 * Retrieve the vdev creation time.
599 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
603 * If we're a top-level vdev, try to load the allocation parameters.
605 if (parent && !parent->vdev_parent &&
606 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
607 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
609 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
611 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
613 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
615 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
618 ASSERT0(vd->vdev_top_zap);
621 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
622 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
623 alloctype == VDEV_ALLOC_ADD ||
624 alloctype == VDEV_ALLOC_SPLIT ||
625 alloctype == VDEV_ALLOC_ROOTPOOL);
626 vd->vdev_mg = metaslab_group_create(islog ?
627 spa_log_class(spa) : spa_normal_class(spa), vd);
630 if (vd->vdev_ops->vdev_op_leaf &&
631 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
632 (void) nvlist_lookup_uint64(nv,
633 ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
635 ASSERT0(vd->vdev_leaf_zap);
639 * If we're a leaf vdev, try to load the DTL object and other state.
642 if (vd->vdev_ops->vdev_op_leaf &&
643 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
644 alloctype == VDEV_ALLOC_ROOTPOOL)) {
645 if (alloctype == VDEV_ALLOC_LOAD) {
646 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
647 &vd->vdev_dtl_object);
648 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
652 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
655 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
656 &spare) == 0 && spare)
660 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
663 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
664 &vd->vdev_resilver_txg);
667 * When importing a pool, we want to ignore the persistent fault
668 * state, as the diagnosis made on another system may not be
669 * valid in the current context. Local vdevs will
670 * remain in the faulted state.
672 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
673 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
675 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
677 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
680 if (vd->vdev_faulted || vd->vdev_degraded) {
684 VDEV_AUX_ERR_EXCEEDED;
685 if (nvlist_lookup_string(nv,
686 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
687 strcmp(aux, "external") == 0)
688 vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
694 * Add ourselves to the parent's list of children.
696 vdev_add_child(parent, vd);
704 vdev_free(vdev_t *vd)
706 spa_t *spa = vd->vdev_spa;
709 * vdev_free() implies closing the vdev first. This is simpler than
710 * trying to ensure complicated semantics for all callers.
714 ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
715 ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
720 for (int c = 0; c < vd->vdev_children; c++)
721 vdev_free(vd->vdev_child[c]);
723 ASSERT(vd->vdev_child == NULL);
724 ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
727 * Discard allocation state.
729 if (vd->vdev_mg != NULL) {
730 vdev_metaslab_fini(vd);
731 metaslab_group_destroy(vd->vdev_mg);
734 ASSERT0(vd->vdev_stat.vs_space);
735 ASSERT0(vd->vdev_stat.vs_dspace);
736 ASSERT0(vd->vdev_stat.vs_alloc);
739 * Remove this vdev from its parent's child list.
741 vdev_remove_child(vd->vdev_parent, vd);
743 ASSERT(vd->vdev_parent == NULL);
746 * Clean up vdev structure.
752 spa_strfree(vd->vdev_path);
754 spa_strfree(vd->vdev_devid);
755 if (vd->vdev_physpath)
756 spa_strfree(vd->vdev_physpath);
758 spa_strfree(vd->vdev_fru);
760 if (vd->vdev_isspare)
761 spa_spare_remove(vd);
762 if (vd->vdev_isl2cache)
763 spa_l2cache_remove(vd);
765 txg_list_destroy(&vd->vdev_ms_list);
766 txg_list_destroy(&vd->vdev_dtl_list);
768 mutex_enter(&vd->vdev_dtl_lock);
769 space_map_close(vd->vdev_dtl_sm);
770 for (int t = 0; t < DTL_TYPES; t++) {
771 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
772 range_tree_destroy(vd->vdev_dtl[t]);
774 mutex_exit(&vd->vdev_dtl_lock);
776 mutex_destroy(&vd->vdev_queue_lock);
777 mutex_destroy(&vd->vdev_dtl_lock);
778 mutex_destroy(&vd->vdev_stat_lock);
779 mutex_destroy(&vd->vdev_probe_lock);
781 if (vd == spa->spa_root_vdev)
782 spa->spa_root_vdev = NULL;
784 kmem_free(vd, sizeof (vdev_t));
788 * Transfer top-level vdev state from svd to tvd.
791 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
793 spa_t *spa = svd->vdev_spa;
798 ASSERT(tvd == tvd->vdev_top);
800 tvd->vdev_ms_array = svd->vdev_ms_array;
801 tvd->vdev_ms_shift = svd->vdev_ms_shift;
802 tvd->vdev_ms_count = svd->vdev_ms_count;
803 tvd->vdev_top_zap = svd->vdev_top_zap;
805 svd->vdev_ms_array = 0;
806 svd->vdev_ms_shift = 0;
807 svd->vdev_ms_count = 0;
808 svd->vdev_top_zap = 0;
811 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
812 tvd->vdev_mg = svd->vdev_mg;
813 tvd->vdev_ms = svd->vdev_ms;
818 if (tvd->vdev_mg != NULL)
819 tvd->vdev_mg->mg_vd = tvd;
821 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
822 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
823 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
825 svd->vdev_stat.vs_alloc = 0;
826 svd->vdev_stat.vs_space = 0;
827 svd->vdev_stat.vs_dspace = 0;
829 for (t = 0; t < TXG_SIZE; t++) {
830 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
831 (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
832 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
833 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
834 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
835 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
838 if (list_link_active(&svd->vdev_config_dirty_node)) {
839 vdev_config_clean(svd);
840 vdev_config_dirty(tvd);
843 if (list_link_active(&svd->vdev_state_dirty_node)) {
844 vdev_state_clean(svd);
845 vdev_state_dirty(tvd);
848 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
849 svd->vdev_deflate_ratio = 0;
851 tvd->vdev_islog = svd->vdev_islog;
856 vdev_top_update(vdev_t *tvd, vdev_t *vd)
863 for (int c = 0; c < vd->vdev_children; c++)
864 vdev_top_update(tvd, vd->vdev_child[c]);
868 * Add a mirror/replacing vdev above an existing vdev.
871 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
873 spa_t *spa = cvd->vdev_spa;
874 vdev_t *pvd = cvd->vdev_parent;
877 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
879 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
881 mvd->vdev_asize = cvd->vdev_asize;
882 mvd->vdev_min_asize = cvd->vdev_min_asize;
883 mvd->vdev_max_asize = cvd->vdev_max_asize;
884 mvd->vdev_ashift = cvd->vdev_ashift;
885 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift;
886 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift;
887 mvd->vdev_state = cvd->vdev_state;
888 mvd->vdev_crtxg = cvd->vdev_crtxg;
890 vdev_remove_child(pvd, cvd);
891 vdev_add_child(pvd, mvd);
892 cvd->vdev_id = mvd->vdev_children;
893 vdev_add_child(mvd, cvd);
894 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
896 if (mvd == mvd->vdev_top)
897 vdev_top_transfer(cvd, mvd);
903 * Remove a 1-way mirror/replacing vdev from the tree.
906 vdev_remove_parent(vdev_t *cvd)
908 vdev_t *mvd = cvd->vdev_parent;
909 vdev_t *pvd = mvd->vdev_parent;
911 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
913 ASSERT(mvd->vdev_children == 1);
914 ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
915 mvd->vdev_ops == &vdev_replacing_ops ||
916 mvd->vdev_ops == &vdev_spare_ops);
917 cvd->vdev_ashift = mvd->vdev_ashift;
918 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift;
919 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift;
921 vdev_remove_child(mvd, cvd);
922 vdev_remove_child(pvd, mvd);
925 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
926 * Otherwise, we could have detached an offline device, and when we
927 * go to import the pool we'll think we have two top-level vdevs,
928 * instead of a different version of the same top-level vdev.
930 if (mvd->vdev_top == mvd) {
931 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
932 cvd->vdev_orig_guid = cvd->vdev_guid;
933 cvd->vdev_guid += guid_delta;
934 cvd->vdev_guid_sum += guid_delta;
936 cvd->vdev_id = mvd->vdev_id;
937 vdev_add_child(pvd, cvd);
938 vdev_top_update(cvd->vdev_top, cvd->vdev_top);
940 if (cvd == cvd->vdev_top)
941 vdev_top_transfer(mvd, cvd);
943 ASSERT(mvd->vdev_children == 0);
948 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
950 spa_t *spa = vd->vdev_spa;
951 objset_t *mos = spa->spa_meta_objset;
953 uint64_t oldc = vd->vdev_ms_count;
954 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
958 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
961 * This vdev is not being allocated from yet or is a hole.
963 if (vd->vdev_ms_shift == 0)
966 ASSERT(!vd->vdev_ishole);
969 * Compute the raidz-deflation ratio. Note, we hard-code
970 * in 128k (1 << 17) because it is the "typical" blocksize.
971 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
972 * otherwise it would inconsistently account for existing bp's.
974 vd->vdev_deflate_ratio = (1 << 17) /
975 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
977 ASSERT(oldc <= newc);
979 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
982 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
983 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
987 vd->vdev_ms_count = newc;
989 for (m = oldc; m < newc; m++) {
993 error = dmu_read(mos, vd->vdev_ms_array,
994 m * sizeof (uint64_t), sizeof (uint64_t), &object,
1000 error = metaslab_init(vd->vdev_mg, m, object, txg,
1007 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
1010 * If the vdev is being removed we don't activate
1011 * the metaslabs since we want to ensure that no new
1012 * allocations are performed on this device.
1014 if (oldc == 0 && !vd->vdev_removing)
1015 metaslab_group_activate(vd->vdev_mg);
1018 spa_config_exit(spa, SCL_ALLOC, FTAG);
1024 vdev_metaslab_fini(vdev_t *vd)
1027 uint64_t count = vd->vdev_ms_count;
1029 if (vd->vdev_ms != NULL) {
1030 metaslab_group_passivate(vd->vdev_mg);
1031 for (m = 0; m < count; m++) {
1032 metaslab_t *msp = vd->vdev_ms[m];
1037 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
1042 typedef struct vdev_probe_stats {
1043 boolean_t vps_readable;
1044 boolean_t vps_writeable;
1046 } vdev_probe_stats_t;
1049 vdev_probe_done(zio_t *zio)
1051 spa_t *spa = zio->io_spa;
1052 vdev_t *vd = zio->io_vd;
1053 vdev_probe_stats_t *vps = zio->io_private;
1055 ASSERT(vd->vdev_probe_zio != NULL);
1057 if (zio->io_type == ZIO_TYPE_READ) {
1058 if (zio->io_error == 0)
1059 vps->vps_readable = 1;
1060 if (zio->io_error == 0 && spa_writeable(spa)) {
1061 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
1062 zio->io_offset, zio->io_size, zio->io_data,
1063 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1064 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
1066 zio_buf_free(zio->io_data, zio->io_size);
1068 } else if (zio->io_type == ZIO_TYPE_WRITE) {
1069 if (zio->io_error == 0)
1070 vps->vps_writeable = 1;
1071 zio_buf_free(zio->io_data, zio->io_size);
1072 } else if (zio->io_type == ZIO_TYPE_NULL) {
1075 vd->vdev_cant_read |= !vps->vps_readable;
1076 vd->vdev_cant_write |= !vps->vps_writeable;
1078 if (vdev_readable(vd) &&
1079 (vdev_writeable(vd) || !spa_writeable(spa))) {
1082 ASSERT(zio->io_error != 0);
1083 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
1084 spa, vd, NULL, 0, 0);
1085 zio->io_error = SET_ERROR(ENXIO);
1088 mutex_enter(&vd->vdev_probe_lock);
1089 ASSERT(vd->vdev_probe_zio == zio);
1090 vd->vdev_probe_zio = NULL;
1091 mutex_exit(&vd->vdev_probe_lock);
1093 zio_link_t *zl = NULL;
1094 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
1095 if (!vdev_accessible(vd, pio))
1096 pio->io_error = SET_ERROR(ENXIO);
1098 kmem_free(vps, sizeof (*vps));
1103 * Determine whether this device is accessible.
1105 * Read and write to several known locations: the pad regions of each
1106 * vdev label but the first, which we leave alone in case it contains
1110 vdev_probe(vdev_t *vd, zio_t *zio)
1112 spa_t *spa = vd->vdev_spa;
1113 vdev_probe_stats_t *vps = NULL;
1116 ASSERT(vd->vdev_ops->vdev_op_leaf);
1119 * Don't probe the probe.
1121 if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
1125 * To prevent 'probe storms' when a device fails, we create
1126 * just one probe i/o at a time. All zios that want to probe
1127 * this vdev will become parents of the probe io.
1129 mutex_enter(&vd->vdev_probe_lock);
1131 if ((pio = vd->vdev_probe_zio) == NULL) {
1132 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
1134 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
1135 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
1138 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
1140 * vdev_cant_read and vdev_cant_write can only
1141 * transition from TRUE to FALSE when we have the
1142 * SCL_ZIO lock as writer; otherwise they can only
1143 * transition from FALSE to TRUE. This ensures that
1144 * any zio looking at these values can assume that
1145 * failures persist for the life of the I/O. That's
1146 * important because when a device has intermittent
1147 * connectivity problems, we want to ensure that
1148 * they're ascribed to the device (ENXIO) and not
1151 * Since we hold SCL_ZIO as writer here, clear both
1152 * values so the probe can reevaluate from first
1155 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1156 vd->vdev_cant_read = B_FALSE;
1157 vd->vdev_cant_write = B_FALSE;
1160 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1161 vdev_probe_done, vps,
1162 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1165 * We can't change the vdev state in this context, so we
1166 * kick off an async task to do it on our behalf.
1169 vd->vdev_probe_wanted = B_TRUE;
1170 spa_async_request(spa, SPA_ASYNC_PROBE);
1175 zio_add_child(zio, pio);
1177 mutex_exit(&vd->vdev_probe_lock);
1180 ASSERT(zio != NULL);
1184 for (int l = 1; l < VDEV_LABELS; l++) {
1185 zio_nowait(zio_read_phys(pio, vd,
1186 vdev_label_offset(vd->vdev_psize, l,
1187 offsetof(vdev_label_t, vl_pad2)),
1188 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1189 ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1190 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1201 vdev_open_child(void *arg)
1205 vd->vdev_open_thread = curthread;
1206 vd->vdev_open_error = vdev_open(vd);
1207 vd->vdev_open_thread = NULL;
1211 vdev_uses_zvols(vdev_t *vd)
1213 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1214 strlen(ZVOL_DIR)) == 0)
1216 for (int c = 0; c < vd->vdev_children; c++)
1217 if (vdev_uses_zvols(vd->vdev_child[c]))
1223 vdev_open_children(vdev_t *vd)
1226 int children = vd->vdev_children;
1229 * in order to handle pools on top of zvols, do the opens
1230 * in a single thread so that the same thread holds the
1231 * spa_namespace_lock
1233 if (B_TRUE || vdev_uses_zvols(vd)) {
1234 for (int c = 0; c < children; c++)
1235 vd->vdev_child[c]->vdev_open_error =
1236 vdev_open(vd->vdev_child[c]);
1239 tq = taskq_create("vdev_open", children, minclsyspri,
1240 children, children, TASKQ_PREPOPULATE);
1242 for (int c = 0; c < children; c++)
1243 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1250 * Prepare a virtual device for access.
1253 vdev_open(vdev_t *vd)
1255 spa_t *spa = vd->vdev_spa;
1258 uint64_t max_osize = 0;
1259 uint64_t asize, max_asize, psize;
1260 uint64_t logical_ashift = 0;
1261 uint64_t physical_ashift = 0;
1263 ASSERT(vd->vdev_open_thread == curthread ||
1264 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1265 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1266 vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1267 vd->vdev_state == VDEV_STATE_OFFLINE);
1269 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1270 vd->vdev_cant_read = B_FALSE;
1271 vd->vdev_cant_write = B_FALSE;
1272 vd->vdev_notrim = B_FALSE;
1273 vd->vdev_min_asize = vdev_get_min_asize(vd);
1276 * If this vdev is not removed, check its fault status. If it's
1277 * faulted, bail out of the open.
1279 if (!vd->vdev_removed && vd->vdev_faulted) {
1280 ASSERT(vd->vdev_children == 0);
1281 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1282 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1283 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1284 vd->vdev_label_aux);
1285 return (SET_ERROR(ENXIO));
1286 } else if (vd->vdev_offline) {
1287 ASSERT(vd->vdev_children == 0);
1288 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1289 return (SET_ERROR(ENXIO));
1292 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize,
1293 &logical_ashift, &physical_ashift);
1296 * Reset the vdev_reopening flag so that we actually close
1297 * the vdev on error.
1299 vd->vdev_reopening = B_FALSE;
1300 if (zio_injection_enabled && error == 0)
1301 error = zio_handle_device_injection(vd, NULL, ENXIO);
1304 if (vd->vdev_removed &&
1305 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1306 vd->vdev_removed = B_FALSE;
1308 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1309 vd->vdev_stat.vs_aux);
1313 vd->vdev_removed = B_FALSE;
1316 * Recheck the faulted flag now that we have confirmed that
1317 * the vdev is accessible. If we're faulted, bail.
1319 if (vd->vdev_faulted) {
1320 ASSERT(vd->vdev_children == 0);
1321 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1322 vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1323 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1324 vd->vdev_label_aux);
1325 return (SET_ERROR(ENXIO));
1328 if (vd->vdev_degraded) {
1329 ASSERT(vd->vdev_children == 0);
1330 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1331 VDEV_AUX_ERR_EXCEEDED);
1333 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1337 * For hole or missing vdevs we just return success.
1339 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1342 if (zfs_trim_enabled && !vd->vdev_notrim && vd->vdev_ops->vdev_op_leaf)
1343 trim_map_create(vd);
1345 for (int c = 0; c < vd->vdev_children; c++) {
1346 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1347 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1353 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1354 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1356 if (vd->vdev_children == 0) {
1357 if (osize < SPA_MINDEVSIZE) {
1358 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1359 VDEV_AUX_TOO_SMALL);
1360 return (SET_ERROR(EOVERFLOW));
1363 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1364 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1365 VDEV_LABEL_END_SIZE);
1367 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1368 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1369 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1370 VDEV_AUX_TOO_SMALL);
1371 return (SET_ERROR(EOVERFLOW));
1375 max_asize = max_osize;
1378 vd->vdev_psize = psize;
1381 * Make sure the allocatable size hasn't shrunk too much.
1383 if (asize < vd->vdev_min_asize) {
1384 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1385 VDEV_AUX_BAD_LABEL);
1386 return (SET_ERROR(EINVAL));
1389 vd->vdev_physical_ashift =
1390 MAX(physical_ashift, vd->vdev_physical_ashift);
1391 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift);
1392 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift);
1394 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) {
1395 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1396 VDEV_AUX_ASHIFT_TOO_BIG);
1400 if (vd->vdev_asize == 0) {
1402 * This is the first-ever open, so use the computed values.
1403 * For testing purposes, a higher ashift can be requested.
1405 vd->vdev_asize = asize;
1406 vd->vdev_max_asize = max_asize;
1409 * Make sure the alignment requirement hasn't increased.
1411 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift &&
1412 vd->vdev_ops->vdev_op_leaf) {
1413 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1414 VDEV_AUX_BAD_LABEL);
1417 vd->vdev_max_asize = max_asize;
1421 * If all children are healthy we update asize if either:
1422 * The asize has increased, due to a device expansion caused by dynamic
1423 * LUN growth or vdev replacement, and automatic expansion is enabled;
1424 * making the additional space available.
1426 * The asize has decreased, due to a device shrink usually caused by a
1427 * vdev replace with a smaller device. This ensures that calculations
1428 * based of max_asize and asize e.g. esize are always valid. It's safe
1429 * to do this as we've already validated that asize is greater than
1432 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1433 ((asize > vd->vdev_asize &&
1434 (vd->vdev_expanding || spa->spa_autoexpand)) ||
1435 (asize < vd->vdev_asize)))
1436 vd->vdev_asize = asize;
1438 vdev_set_min_asize(vd);
1441 * Ensure we can issue some IO before declaring the
1442 * vdev open for business.
1444 if (vd->vdev_ops->vdev_op_leaf &&
1445 (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1446 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1447 VDEV_AUX_ERR_EXCEEDED);
1452 * Track the min and max ashift values for normal data devices.
1454 if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
1455 !vd->vdev_islog && vd->vdev_aux == NULL) {
1456 if (vd->vdev_ashift > spa->spa_max_ashift)
1457 spa->spa_max_ashift = vd->vdev_ashift;
1458 if (vd->vdev_ashift < spa->spa_min_ashift)
1459 spa->spa_min_ashift = vd->vdev_ashift;
1463 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1464 * resilver. But don't do this if we are doing a reopen for a scrub,
1465 * since this would just restart the scrub we are already doing.
1467 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1468 vdev_resilver_needed(vd, NULL, NULL))
1469 spa_async_request(spa, SPA_ASYNC_RESILVER);
1475 * Called once the vdevs are all opened, this routine validates the label
1476 * contents. This needs to be done before vdev_load() so that we don't
1477 * inadvertently do repair I/Os to the wrong device.
1479 * If 'strict' is false ignore the spa guid check. This is necessary because
1480 * if the machine crashed during a re-guid the new guid might have been written
1481 * to all of the vdev labels, but not the cached config. The strict check
1482 * will be performed when the pool is opened again using the mos config.
1484 * This function will only return failure if one of the vdevs indicates that it
1485 * has since been destroyed or exported. This is only possible if
1486 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1487 * will be updated but the function will return 0.
1490 vdev_validate(vdev_t *vd, boolean_t strict)
1492 spa_t *spa = vd->vdev_spa;
1494 uint64_t guid = 0, top_guid;
1497 for (int c = 0; c < vd->vdev_children; c++)
1498 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1499 return (SET_ERROR(EBADF));
1502 * If the device has already failed, or was marked offline, don't do
1503 * any further validation. Otherwise, label I/O will fail and we will
1504 * overwrite the previous state.
1506 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1507 uint64_t aux_guid = 0;
1509 uint64_t txg = spa_last_synced_txg(spa) != 0 ?
1510 spa_last_synced_txg(spa) : -1ULL;
1512 if ((label = vdev_label_read_config(vd, txg)) == NULL) {
1513 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1514 VDEV_AUX_BAD_LABEL);
1519 * Determine if this vdev has been split off into another
1520 * pool. If so, then refuse to open it.
1522 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1523 &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1524 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1525 VDEV_AUX_SPLIT_POOL);
1530 if (strict && (nvlist_lookup_uint64(label,
1531 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1532 guid != spa_guid(spa))) {
1533 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1534 VDEV_AUX_CORRUPT_DATA);
1539 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1540 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1545 * If this vdev just became a top-level vdev because its
1546 * sibling was detached, it will have adopted the parent's
1547 * vdev guid -- but the label may or may not be on disk yet.
1548 * Fortunately, either version of the label will have the
1549 * same top guid, so if we're a top-level vdev, we can
1550 * safely compare to that instead.
1552 * If we split this vdev off instead, then we also check the
1553 * original pool's guid. We don't want to consider the vdev
1554 * corrupt if it is partway through a split operation.
1556 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1558 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1560 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1561 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1562 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1563 VDEV_AUX_CORRUPT_DATA);
1568 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1570 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1571 VDEV_AUX_CORRUPT_DATA);
1579 * If this is a verbatim import, no need to check the
1580 * state of the pool.
1582 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1583 spa_load_state(spa) == SPA_LOAD_OPEN &&
1584 state != POOL_STATE_ACTIVE)
1585 return (SET_ERROR(EBADF));
1588 * If we were able to open and validate a vdev that was
1589 * previously marked permanently unavailable, clear that state
1592 if (vd->vdev_not_present)
1593 vd->vdev_not_present = 0;
1600 * Close a virtual device.
1603 vdev_close(vdev_t *vd)
1605 spa_t *spa = vd->vdev_spa;
1606 vdev_t *pvd = vd->vdev_parent;
1608 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1611 * If our parent is reopening, then we are as well, unless we are
1614 if (pvd != NULL && pvd->vdev_reopening)
1615 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1617 vd->vdev_ops->vdev_op_close(vd);
1619 vdev_cache_purge(vd);
1621 if (vd->vdev_ops->vdev_op_leaf)
1622 trim_map_destroy(vd);
1625 * We record the previous state before we close it, so that if we are
1626 * doing a reopen(), we don't generate FMA ereports if we notice that
1627 * it's still faulted.
1629 vd->vdev_prevstate = vd->vdev_state;
1631 if (vd->vdev_offline)
1632 vd->vdev_state = VDEV_STATE_OFFLINE;
1634 vd->vdev_state = VDEV_STATE_CLOSED;
1635 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1639 vdev_hold(vdev_t *vd)
1641 spa_t *spa = vd->vdev_spa;
1643 ASSERT(spa_is_root(spa));
1644 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1647 for (int c = 0; c < vd->vdev_children; c++)
1648 vdev_hold(vd->vdev_child[c]);
1650 if (vd->vdev_ops->vdev_op_leaf)
1651 vd->vdev_ops->vdev_op_hold(vd);
1655 vdev_rele(vdev_t *vd)
1657 spa_t *spa = vd->vdev_spa;
1659 ASSERT(spa_is_root(spa));
1660 for (int c = 0; c < vd->vdev_children; c++)
1661 vdev_rele(vd->vdev_child[c]);
1663 if (vd->vdev_ops->vdev_op_leaf)
1664 vd->vdev_ops->vdev_op_rele(vd);
1668 * Reopen all interior vdevs and any unopened leaves. We don't actually
1669 * reopen leaf vdevs which had previously been opened as they might deadlock
1670 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
1671 * If the leaf has never been opened then open it, as usual.
1674 vdev_reopen(vdev_t *vd)
1676 spa_t *spa = vd->vdev_spa;
1678 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1680 /* set the reopening flag unless we're taking the vdev offline */
1681 vd->vdev_reopening = !vd->vdev_offline;
1683 (void) vdev_open(vd);
1686 * Call vdev_validate() here to make sure we have the same device.
1687 * Otherwise, a device with an invalid label could be successfully
1688 * opened in response to vdev_reopen().
1691 (void) vdev_validate_aux(vd);
1692 if (vdev_readable(vd) && vdev_writeable(vd) &&
1693 vd->vdev_aux == &spa->spa_l2cache &&
1694 !l2arc_vdev_present(vd))
1695 l2arc_add_vdev(spa, vd);
1697 (void) vdev_validate(vd, B_TRUE);
1701 * Reassess parent vdev's health.
1703 vdev_propagate_state(vd);
1707 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1712 * Normally, partial opens (e.g. of a mirror) are allowed.
1713 * For a create, however, we want to fail the request if
1714 * there are any components we can't open.
1716 error = vdev_open(vd);
1718 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1720 return (error ? error : ENXIO);
1724 * Recursively load DTLs and initialize all labels.
1726 if ((error = vdev_dtl_load(vd)) != 0 ||
1727 (error = vdev_label_init(vd, txg, isreplacing ?
1728 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1737 vdev_metaslab_set_size(vdev_t *vd)
1740 * Aim for roughly metaslabs_per_vdev (default 200) metaslabs per vdev.
1742 vd->vdev_ms_shift = highbit64(vd->vdev_asize / metaslabs_per_vdev);
1743 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1747 * Maximize performance by inflating the configured ashift for top level
1748 * vdevs to be as close to the physical ashift as possible while maintaining
1749 * administrator defined limits and ensuring it doesn't go below the
1753 vdev_ashift_optimize(vdev_t *vd)
1755 if (vd == vd->vdev_top) {
1756 if (vd->vdev_ashift < vd->vdev_physical_ashift) {
1757 vd->vdev_ashift = MIN(
1758 MAX(zfs_max_auto_ashift, vd->vdev_ashift),
1759 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift));
1762 * Unusual case where logical ashift > physical ashift
1763 * so we can't cap the calculated ashift based on max
1764 * ashift as that would cause failures.
1765 * We still check if we need to increase it to match
1768 vd->vdev_ashift = MAX(zfs_min_auto_ashift,
1775 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1777 ASSERT(vd == vd->vdev_top);
1778 ASSERT(!vd->vdev_ishole);
1779 ASSERT(ISP2(flags));
1780 ASSERT(spa_writeable(vd->vdev_spa));
1782 if (flags & VDD_METASLAB)
1783 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1785 if (flags & VDD_DTL)
1786 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1788 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1792 vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
1794 for (int c = 0; c < vd->vdev_children; c++)
1795 vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
1797 if (vd->vdev_ops->vdev_op_leaf)
1798 vdev_dirty(vd->vdev_top, flags, vd, txg);
1804 * A vdev's DTL (dirty time log) is the set of transaction groups for which
1805 * the vdev has less than perfect replication. There are four kinds of DTL:
1807 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1809 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1811 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1812 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1813 * txgs that was scrubbed.
1815 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1816 * persistent errors or just some device being offline.
1817 * Unlike the other three, the DTL_OUTAGE map is not generally
1818 * maintained; it's only computed when needed, typically to
1819 * determine whether a device can be detached.
1821 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1822 * either has the data or it doesn't.
1824 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1825 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1826 * if any child is less than fully replicated, then so is its parent.
1827 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1828 * comprising only those txgs which appear in 'maxfaults' or more children;
1829 * those are the txgs we don't have enough replication to read. For example,
1830 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1831 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1832 * two child DTL_MISSING maps.
1834 * It should be clear from the above that to compute the DTLs and outage maps
1835 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1836 * Therefore, that is all we keep on disk. When loading the pool, or after
1837 * a configuration change, we generate all other DTLs from first principles.
1840 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1842 range_tree_t *rt = vd->vdev_dtl[t];
1844 ASSERT(t < DTL_TYPES);
1845 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1846 ASSERT(spa_writeable(vd->vdev_spa));
1848 mutex_enter(rt->rt_lock);
1849 if (!range_tree_contains(rt, txg, size))
1850 range_tree_add(rt, txg, size);
1851 mutex_exit(rt->rt_lock);
1855 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1857 range_tree_t *rt = vd->vdev_dtl[t];
1858 boolean_t dirty = B_FALSE;
1860 ASSERT(t < DTL_TYPES);
1861 ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1863 mutex_enter(rt->rt_lock);
1864 if (range_tree_space(rt) != 0)
1865 dirty = range_tree_contains(rt, txg, size);
1866 mutex_exit(rt->rt_lock);
1872 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1874 range_tree_t *rt = vd->vdev_dtl[t];
1877 mutex_enter(rt->rt_lock);
1878 empty = (range_tree_space(rt) == 0);
1879 mutex_exit(rt->rt_lock);
1885 * Returns the lowest txg in the DTL range.
1888 vdev_dtl_min(vdev_t *vd)
1892 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1893 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1894 ASSERT0(vd->vdev_children);
1896 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1897 return (rs->rs_start - 1);
1901 * Returns the highest txg in the DTL.
1904 vdev_dtl_max(vdev_t *vd)
1908 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
1909 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
1910 ASSERT0(vd->vdev_children);
1912 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
1913 return (rs->rs_end);
1917 * Determine if a resilvering vdev should remove any DTL entries from
1918 * its range. If the vdev was resilvering for the entire duration of the
1919 * scan then it should excise that range from its DTLs. Otherwise, this
1920 * vdev is considered partially resilvered and should leave its DTL
1921 * entries intact. The comment in vdev_dtl_reassess() describes how we
1925 vdev_dtl_should_excise(vdev_t *vd)
1927 spa_t *spa = vd->vdev_spa;
1928 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1930 ASSERT0(scn->scn_phys.scn_errors);
1931 ASSERT0(vd->vdev_children);
1933 if (vd->vdev_state < VDEV_STATE_DEGRADED)
1936 if (vd->vdev_resilver_txg == 0 ||
1937 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
1941 * When a resilver is initiated the scan will assign the scn_max_txg
1942 * value to the highest txg value that exists in all DTLs. If this
1943 * device's max DTL is not part of this scan (i.e. it is not in
1944 * the range (scn_min_txg, scn_max_txg] then it is not eligible
1947 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) {
1948 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd));
1949 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg);
1950 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg);
1957 * Reassess DTLs after a config change or scrub completion.
1960 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1962 spa_t *spa = vd->vdev_spa;
1966 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1968 for (int c = 0; c < vd->vdev_children; c++)
1969 vdev_dtl_reassess(vd->vdev_child[c], txg,
1970 scrub_txg, scrub_done);
1972 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1975 if (vd->vdev_ops->vdev_op_leaf) {
1976 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1978 mutex_enter(&vd->vdev_dtl_lock);
1981 * If we've completed a scan cleanly then determine
1982 * if this vdev should remove any DTLs. We only want to
1983 * excise regions on vdevs that were available during
1984 * the entire duration of this scan.
1986 if (scrub_txg != 0 &&
1987 (spa->spa_scrub_started ||
1988 (scn != NULL && scn->scn_phys.scn_errors == 0)) &&
1989 vdev_dtl_should_excise(vd)) {
1991 * We completed a scrub up to scrub_txg. If we
1992 * did it without rebooting, then the scrub dtl
1993 * will be valid, so excise the old region and
1994 * fold in the scrub dtl. Otherwise, leave the
1995 * dtl as-is if there was an error.
1997 * There's little trick here: to excise the beginning
1998 * of the DTL_MISSING map, we put it into a reference
1999 * tree and then add a segment with refcnt -1 that
2000 * covers the range [0, scrub_txg). This means
2001 * that each txg in that range has refcnt -1 or 0.
2002 * We then add DTL_SCRUB with a refcnt of 2, so that
2003 * entries in the range [0, scrub_txg) will have a
2004 * positive refcnt -- either 1 or 2. We then convert
2005 * the reference tree into the new DTL_MISSING map.
2007 space_reftree_create(&reftree);
2008 space_reftree_add_map(&reftree,
2009 vd->vdev_dtl[DTL_MISSING], 1);
2010 space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
2011 space_reftree_add_map(&reftree,
2012 vd->vdev_dtl[DTL_SCRUB], 2);
2013 space_reftree_generate_map(&reftree,
2014 vd->vdev_dtl[DTL_MISSING], 1);
2015 space_reftree_destroy(&reftree);
2017 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
2018 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2019 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
2021 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
2022 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
2023 if (!vdev_readable(vd))
2024 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
2026 range_tree_walk(vd->vdev_dtl[DTL_MISSING],
2027 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
2030 * If the vdev was resilvering and no longer has any
2031 * DTLs then reset its resilvering flag and dirty
2032 * the top level so that we persist the change.
2034 if (vd->vdev_resilver_txg != 0 &&
2035 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
2036 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) {
2037 vd->vdev_resilver_txg = 0;
2038 vdev_config_dirty(vd->vdev_top);
2041 mutex_exit(&vd->vdev_dtl_lock);
2044 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
2048 mutex_enter(&vd->vdev_dtl_lock);
2049 for (int t = 0; t < DTL_TYPES; t++) {
2050 /* account for child's outage in parent's missing map */
2051 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
2053 continue; /* leaf vdevs only */
2054 if (t == DTL_PARTIAL)
2055 minref = 1; /* i.e. non-zero */
2056 else if (vd->vdev_nparity != 0)
2057 minref = vd->vdev_nparity + 1; /* RAID-Z */
2059 minref = vd->vdev_children; /* any kind of mirror */
2060 space_reftree_create(&reftree);
2061 for (int c = 0; c < vd->vdev_children; c++) {
2062 vdev_t *cvd = vd->vdev_child[c];
2063 mutex_enter(&cvd->vdev_dtl_lock);
2064 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
2065 mutex_exit(&cvd->vdev_dtl_lock);
2067 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
2068 space_reftree_destroy(&reftree);
2070 mutex_exit(&vd->vdev_dtl_lock);
2074 vdev_dtl_load(vdev_t *vd)
2076 spa_t *spa = vd->vdev_spa;
2077 objset_t *mos = spa->spa_meta_objset;
2080 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
2081 ASSERT(!vd->vdev_ishole);
2083 error = space_map_open(&vd->vdev_dtl_sm, mos,
2084 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
2087 ASSERT(vd->vdev_dtl_sm != NULL);
2089 mutex_enter(&vd->vdev_dtl_lock);
2092 * Now that we've opened the space_map we need to update
2095 space_map_update(vd->vdev_dtl_sm);
2097 error = space_map_load(vd->vdev_dtl_sm,
2098 vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
2099 mutex_exit(&vd->vdev_dtl_lock);
2104 for (int c = 0; c < vd->vdev_children; c++) {
2105 error = vdev_dtl_load(vd->vdev_child[c]);
2114 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
2116 spa_t *spa = vd->vdev_spa;
2118 VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
2119 VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2124 vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
2126 spa_t *spa = vd->vdev_spa;
2127 uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
2128 DMU_OT_NONE, 0, tx);
2131 VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
2138 vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
2140 if (vd->vdev_ops != &vdev_hole_ops &&
2141 vd->vdev_ops != &vdev_missing_ops &&
2142 vd->vdev_ops != &vdev_root_ops &&
2143 !vd->vdev_top->vdev_removing) {
2144 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
2145 vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
2147 if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
2148 vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
2151 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2152 vdev_construct_zaps(vd->vdev_child[i], tx);
2157 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
2159 spa_t *spa = vd->vdev_spa;
2160 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
2161 objset_t *mos = spa->spa_meta_objset;
2162 range_tree_t *rtsync;
2165 uint64_t object = space_map_object(vd->vdev_dtl_sm);
2167 ASSERT(!vd->vdev_ishole);
2168 ASSERT(vd->vdev_ops->vdev_op_leaf);
2170 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2172 if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
2173 mutex_enter(&vd->vdev_dtl_lock);
2174 space_map_free(vd->vdev_dtl_sm, tx);
2175 space_map_close(vd->vdev_dtl_sm);
2176 vd->vdev_dtl_sm = NULL;
2177 mutex_exit(&vd->vdev_dtl_lock);
2180 * We only destroy the leaf ZAP for detached leaves or for
2181 * removed log devices. Removed data devices handle leaf ZAP
2182 * cleanup later, once cancellation is no longer possible.
2184 if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
2185 vd->vdev_top->vdev_islog)) {
2186 vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
2187 vd->vdev_leaf_zap = 0;
2194 if (vd->vdev_dtl_sm == NULL) {
2195 uint64_t new_object;
2197 new_object = space_map_alloc(mos, tx);
2198 VERIFY3U(new_object, !=, 0);
2200 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
2201 0, -1ULL, 0, &vd->vdev_dtl_lock));
2202 ASSERT(vd->vdev_dtl_sm != NULL);
2205 bzero(&rtlock, sizeof(rtlock));
2206 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
2208 rtsync = range_tree_create(NULL, NULL, &rtlock);
2210 mutex_enter(&rtlock);
2212 mutex_enter(&vd->vdev_dtl_lock);
2213 range_tree_walk(rt, range_tree_add, rtsync);
2214 mutex_exit(&vd->vdev_dtl_lock);
2216 space_map_truncate(vd->vdev_dtl_sm, tx);
2217 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
2218 range_tree_vacate(rtsync, NULL, NULL);
2220 range_tree_destroy(rtsync);
2222 mutex_exit(&rtlock);
2223 mutex_destroy(&rtlock);
2226 * If the object for the space map has changed then dirty
2227 * the top level so that we update the config.
2229 if (object != space_map_object(vd->vdev_dtl_sm)) {
2230 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
2231 "new object %llu", txg, spa_name(spa), object,
2232 space_map_object(vd->vdev_dtl_sm));
2233 vdev_config_dirty(vd->vdev_top);
2238 mutex_enter(&vd->vdev_dtl_lock);
2239 space_map_update(vd->vdev_dtl_sm);
2240 mutex_exit(&vd->vdev_dtl_lock);
2244 * Determine whether the specified vdev can be offlined/detached/removed
2245 * without losing data.
2248 vdev_dtl_required(vdev_t *vd)
2250 spa_t *spa = vd->vdev_spa;
2251 vdev_t *tvd = vd->vdev_top;
2252 uint8_t cant_read = vd->vdev_cant_read;
2255 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2257 if (vd == spa->spa_root_vdev || vd == tvd)
2261 * Temporarily mark the device as unreadable, and then determine
2262 * whether this results in any DTL outages in the top-level vdev.
2263 * If not, we can safely offline/detach/remove the device.
2265 vd->vdev_cant_read = B_TRUE;
2266 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2267 required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
2268 vd->vdev_cant_read = cant_read;
2269 vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
2271 if (!required && zio_injection_enabled)
2272 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
2278 * Determine if resilver is needed, and if so the txg range.
2281 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
2283 boolean_t needed = B_FALSE;
2284 uint64_t thismin = UINT64_MAX;
2285 uint64_t thismax = 0;
2287 if (vd->vdev_children == 0) {
2288 mutex_enter(&vd->vdev_dtl_lock);
2289 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
2290 vdev_writeable(vd)) {
2292 thismin = vdev_dtl_min(vd);
2293 thismax = vdev_dtl_max(vd);
2296 mutex_exit(&vd->vdev_dtl_lock);
2298 for (int c = 0; c < vd->vdev_children; c++) {
2299 vdev_t *cvd = vd->vdev_child[c];
2300 uint64_t cmin, cmax;
2302 if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
2303 thismin = MIN(thismin, cmin);
2304 thismax = MAX(thismax, cmax);
2310 if (needed && minp) {
2318 vdev_load(vdev_t *vd)
2321 * Recursively load all children.
2323 for (int c = 0; c < vd->vdev_children; c++)
2324 vdev_load(vd->vdev_child[c]);
2327 * If this is a top-level vdev, initialize its metaslabs.
2329 if (vd == vd->vdev_top && !vd->vdev_ishole &&
2330 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
2331 vdev_metaslab_init(vd, 0) != 0))
2332 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2333 VDEV_AUX_CORRUPT_DATA);
2336 * If this is a leaf vdev, load its DTL.
2338 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
2339 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
2340 VDEV_AUX_CORRUPT_DATA);
2344 * The special vdev case is used for hot spares and l2cache devices. Its
2345 * sole purpose it to set the vdev state for the associated vdev. To do this,
2346 * we make sure that we can open the underlying device, then try to read the
2347 * label, and make sure that the label is sane and that it hasn't been
2348 * repurposed to another pool.
2351 vdev_validate_aux(vdev_t *vd)
2354 uint64_t guid, version;
2357 if (!vdev_readable(vd))
2360 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) {
2361 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2362 VDEV_AUX_CORRUPT_DATA);
2366 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
2367 !SPA_VERSION_IS_SUPPORTED(version) ||
2368 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
2369 guid != vd->vdev_guid ||
2370 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
2371 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
2372 VDEV_AUX_CORRUPT_DATA);
2378 * We don't actually check the pool state here. If it's in fact in
2379 * use by another pool, we update this fact on the fly when requested.
2386 vdev_remove(vdev_t *vd, uint64_t txg)
2388 spa_t *spa = vd->vdev_spa;
2389 objset_t *mos = spa->spa_meta_objset;
2392 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2393 ASSERT(vd == vd->vdev_top);
2394 ASSERT3U(txg, ==, spa_syncing_txg(spa));
2396 if (vd->vdev_ms != NULL) {
2397 metaslab_group_t *mg = vd->vdev_mg;
2399 metaslab_group_histogram_verify(mg);
2400 metaslab_class_histogram_verify(mg->mg_class);
2402 for (int m = 0; m < vd->vdev_ms_count; m++) {
2403 metaslab_t *msp = vd->vdev_ms[m];
2405 if (msp == NULL || msp->ms_sm == NULL)
2408 mutex_enter(&msp->ms_lock);
2410 * If the metaslab was not loaded when the vdev
2411 * was removed then the histogram accounting may
2412 * not be accurate. Update the histogram information
2413 * here so that we ensure that the metaslab group
2414 * and metaslab class are up-to-date.
2416 metaslab_group_histogram_remove(mg, msp);
2418 VERIFY0(space_map_allocated(msp->ms_sm));
2419 space_map_free(msp->ms_sm, tx);
2420 space_map_close(msp->ms_sm);
2422 mutex_exit(&msp->ms_lock);
2425 metaslab_group_histogram_verify(mg);
2426 metaslab_class_histogram_verify(mg->mg_class);
2427 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
2428 ASSERT0(mg->mg_histogram[i]);
2432 if (vd->vdev_ms_array) {
2433 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2434 vd->vdev_ms_array = 0;
2437 if (vd->vdev_islog && vd->vdev_top_zap != 0) {
2438 vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
2439 vd->vdev_top_zap = 0;
2445 vdev_sync_done(vdev_t *vd, uint64_t txg)
2448 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2450 ASSERT(!vd->vdev_ishole);
2452 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2453 metaslab_sync_done(msp, txg);
2456 metaslab_sync_reassess(vd->vdev_mg);
2460 vdev_sync(vdev_t *vd, uint64_t txg)
2462 spa_t *spa = vd->vdev_spa;
2467 ASSERT(!vd->vdev_ishole);
2469 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2470 ASSERT(vd == vd->vdev_top);
2471 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2472 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2473 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2474 ASSERT(vd->vdev_ms_array != 0);
2475 vdev_config_dirty(vd);
2480 * Remove the metadata associated with this vdev once it's empty.
2482 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2483 vdev_remove(vd, txg);
2485 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2486 metaslab_sync(msp, txg);
2487 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2490 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2491 vdev_dtl_sync(lvd, txg);
2493 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2497 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2499 return (vd->vdev_ops->vdev_op_asize(vd, psize));
2503 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
2504 * not be opened, and no I/O is attempted.
2507 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2511 spa_vdev_state_enter(spa, SCL_NONE);
2513 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2514 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2516 if (!vd->vdev_ops->vdev_op_leaf)
2517 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2522 * We don't directly use the aux state here, but if we do a
2523 * vdev_reopen(), we need this value to be present to remember why we
2526 vd->vdev_label_aux = aux;
2529 * Faulted state takes precedence over degraded.
2531 vd->vdev_delayed_close = B_FALSE;
2532 vd->vdev_faulted = 1ULL;
2533 vd->vdev_degraded = 0ULL;
2534 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2537 * If this device has the only valid copy of the data, then
2538 * back off and simply mark the vdev as degraded instead.
2540 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2541 vd->vdev_degraded = 1ULL;
2542 vd->vdev_faulted = 0ULL;
2545 * If we reopen the device and it's not dead, only then do we
2550 if (vdev_readable(vd))
2551 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2554 return (spa_vdev_state_exit(spa, vd, 0));
2558 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
2559 * user that something is wrong. The vdev continues to operate as normal as far
2560 * as I/O is concerned.
2563 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2567 spa_vdev_state_enter(spa, SCL_NONE);
2569 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2570 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2572 if (!vd->vdev_ops->vdev_op_leaf)
2573 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2576 * If the vdev is already faulted, then don't do anything.
2578 if (vd->vdev_faulted || vd->vdev_degraded)
2579 return (spa_vdev_state_exit(spa, NULL, 0));
2581 vd->vdev_degraded = 1ULL;
2582 if (!vdev_is_dead(vd))
2583 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2586 return (spa_vdev_state_exit(spa, vd, 0));
2590 * Online the given vdev.
2592 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
2593 * spare device should be detached when the device finishes resilvering.
2594 * Second, the online should be treated like a 'test' online case, so no FMA
2595 * events are generated if the device fails to open.
2598 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2600 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2601 boolean_t wasoffline;
2602 vdev_state_t oldstate;
2604 spa_vdev_state_enter(spa, SCL_NONE);
2606 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2607 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2609 if (!vd->vdev_ops->vdev_op_leaf)
2610 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2612 wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline);
2613 oldstate = vd->vdev_state;
2616 vd->vdev_offline = B_FALSE;
2617 vd->vdev_tmpoffline = B_FALSE;
2618 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2619 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2621 /* XXX - L2ARC 1.0 does not support expansion */
2622 if (!vd->vdev_aux) {
2623 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2624 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2628 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2630 if (!vd->vdev_aux) {
2631 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2632 pvd->vdev_expanding = B_FALSE;
2636 *newstate = vd->vdev_state;
2637 if ((flags & ZFS_ONLINE_UNSPARE) &&
2638 !vdev_is_dead(vd) && vd->vdev_parent &&
2639 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2640 vd->vdev_parent->vdev_child[0] == vd)
2641 vd->vdev_unspare = B_TRUE;
2643 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2645 /* XXX - L2ARC 1.0 does not support expansion */
2647 return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2648 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2652 (oldstate < VDEV_STATE_DEGRADED &&
2653 vd->vdev_state >= VDEV_STATE_DEGRADED))
2654 spa_event_notify(spa, vd, ESC_ZFS_VDEV_ONLINE);
2656 return (spa_vdev_state_exit(spa, vd, 0));
2660 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2664 uint64_t generation;
2665 metaslab_group_t *mg;
2668 spa_vdev_state_enter(spa, SCL_ALLOC);
2670 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2671 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2673 if (!vd->vdev_ops->vdev_op_leaf)
2674 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2678 generation = spa->spa_config_generation + 1;
2681 * If the device isn't already offline, try to offline it.
2683 if (!vd->vdev_offline) {
2685 * If this device has the only valid copy of some data,
2686 * don't allow it to be offlined. Log devices are always
2689 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2690 vdev_dtl_required(vd))
2691 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2694 * If the top-level is a slog and it has had allocations
2695 * then proceed. We check that the vdev's metaslab group
2696 * is not NULL since it's possible that we may have just
2697 * added this vdev but not yet initialized its metaslabs.
2699 if (tvd->vdev_islog && mg != NULL) {
2701 * Prevent any future allocations.
2703 metaslab_group_passivate(mg);
2704 (void) spa_vdev_state_exit(spa, vd, 0);
2706 error = spa_offline_log(spa);
2708 spa_vdev_state_enter(spa, SCL_ALLOC);
2711 * Check to see if the config has changed.
2713 if (error || generation != spa->spa_config_generation) {
2714 metaslab_group_activate(mg);
2716 return (spa_vdev_state_exit(spa,
2718 (void) spa_vdev_state_exit(spa, vd, 0);
2721 ASSERT0(tvd->vdev_stat.vs_alloc);
2725 * Offline this device and reopen its top-level vdev.
2726 * If the top-level vdev is a log device then just offline
2727 * it. Otherwise, if this action results in the top-level
2728 * vdev becoming unusable, undo it and fail the request.
2730 vd->vdev_offline = B_TRUE;
2733 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2734 vdev_is_dead(tvd)) {
2735 vd->vdev_offline = B_FALSE;
2737 return (spa_vdev_state_exit(spa, NULL, EBUSY));
2741 * Add the device back into the metaslab rotor so that
2742 * once we online the device it's open for business.
2744 if (tvd->vdev_islog && mg != NULL)
2745 metaslab_group_activate(mg);
2748 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2750 return (spa_vdev_state_exit(spa, vd, 0));
2754 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2758 mutex_enter(&spa->spa_vdev_top_lock);
2759 error = vdev_offline_locked(spa, guid, flags);
2760 mutex_exit(&spa->spa_vdev_top_lock);
2766 * Clear the error counts associated with this vdev. Unlike vdev_online() and
2767 * vdev_offline(), we assume the spa config is locked. We also clear all
2768 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
2771 vdev_clear(spa_t *spa, vdev_t *vd)
2773 vdev_t *rvd = spa->spa_root_vdev;
2775 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2780 vd->vdev_stat.vs_read_errors = 0;
2781 vd->vdev_stat.vs_write_errors = 0;
2782 vd->vdev_stat.vs_checksum_errors = 0;
2784 for (int c = 0; c < vd->vdev_children; c++)
2785 vdev_clear(spa, vd->vdev_child[c]);
2788 for (int c = 0; c < spa->spa_l2cache.sav_count; c++)
2789 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]);
2791 for (int c = 0; c < spa->spa_spares.sav_count; c++)
2792 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]);
2796 * If we're in the FAULTED state or have experienced failed I/O, then
2797 * clear the persistent state and attempt to reopen the device. We
2798 * also mark the vdev config dirty, so that the new faulted state is
2799 * written out to disk.
2801 if (vd->vdev_faulted || vd->vdev_degraded ||
2802 !vdev_readable(vd) || !vdev_writeable(vd)) {
2805 * When reopening in reponse to a clear event, it may be due to
2806 * a fmadm repair request. In this case, if the device is
2807 * still broken, we want to still post the ereport again.
2809 vd->vdev_forcefault = B_TRUE;
2811 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2812 vd->vdev_cant_read = B_FALSE;
2813 vd->vdev_cant_write = B_FALSE;
2815 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2817 vd->vdev_forcefault = B_FALSE;
2819 if (vd != rvd && vdev_writeable(vd->vdev_top))
2820 vdev_state_dirty(vd->vdev_top);
2822 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2823 spa_async_request(spa, SPA_ASYNC_RESILVER);
2825 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2829 * When clearing a FMA-diagnosed fault, we always want to
2830 * unspare the device, as we assume that the original spare was
2831 * done in response to the FMA fault.
2833 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2834 vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2835 vd->vdev_parent->vdev_child[0] == vd)
2836 vd->vdev_unspare = B_TRUE;
2840 vdev_is_dead(vdev_t *vd)
2843 * Holes and missing devices are always considered "dead".
2844 * This simplifies the code since we don't have to check for
2845 * these types of devices in the various code paths.
2846 * Instead we rely on the fact that we skip over dead devices
2847 * before issuing I/O to them.
2849 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2850 vd->vdev_ops == &vdev_missing_ops);
2854 vdev_readable(vdev_t *vd)
2856 return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2860 vdev_writeable(vdev_t *vd)
2862 return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2866 vdev_allocatable(vdev_t *vd)
2868 uint64_t state = vd->vdev_state;
2871 * We currently allow allocations from vdevs which may be in the
2872 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2873 * fails to reopen then we'll catch it later when we're holding
2874 * the proper locks. Note that we have to get the vdev state
2875 * in a local variable because although it changes atomically,
2876 * we're asking two separate questions about it.
2878 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2879 !vd->vdev_cant_write && !vd->vdev_ishole &&
2880 vd->vdev_mg->mg_initialized);
2884 vdev_accessible(vdev_t *vd, zio_t *zio)
2886 ASSERT(zio->io_vd == vd);
2888 if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2891 if (zio->io_type == ZIO_TYPE_READ)
2892 return (!vd->vdev_cant_read);
2894 if (zio->io_type == ZIO_TYPE_WRITE)
2895 return (!vd->vdev_cant_write);
2901 * Get statistics for the given vdev.
2904 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2906 spa_t *spa = vd->vdev_spa;
2907 vdev_t *rvd = spa->spa_root_vdev;
2908 vdev_t *tvd = vd->vdev_top;
2910 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2912 mutex_enter(&vd->vdev_stat_lock);
2913 bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2914 vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2915 vs->vs_state = vd->vdev_state;
2916 vs->vs_rsize = vdev_get_min_asize(vd);
2917 if (vd->vdev_ops->vdev_op_leaf)
2918 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2920 * Report expandable space on top-level, non-auxillary devices only.
2921 * The expandable space is reported in terms of metaslab sized units
2922 * since that determines how much space the pool can expand.
2924 if (vd->vdev_aux == NULL && tvd != NULL && vd->vdev_max_asize != 0) {
2925 vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize,
2926 1ULL << tvd->vdev_ms_shift);
2928 vs->vs_configured_ashift = vd->vdev_top != NULL
2929 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift;
2930 vs->vs_logical_ashift = vd->vdev_logical_ashift;
2931 vs->vs_physical_ashift = vd->vdev_physical_ashift;
2932 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) {
2933 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
2937 * If we're getting stats on the root vdev, aggregate the I/O counts
2938 * over all top-level vdevs (i.e. the direct children of the root).
2941 for (int c = 0; c < rvd->vdev_children; c++) {
2942 vdev_t *cvd = rvd->vdev_child[c];
2943 vdev_stat_t *cvs = &cvd->vdev_stat;
2945 for (int t = 0; t < ZIO_TYPES; t++) {
2946 vs->vs_ops[t] += cvs->vs_ops[t];
2947 vs->vs_bytes[t] += cvs->vs_bytes[t];
2949 cvs->vs_scan_removing = cvd->vdev_removing;
2952 mutex_exit(&vd->vdev_stat_lock);
2956 vdev_clear_stats(vdev_t *vd)
2958 mutex_enter(&vd->vdev_stat_lock);
2959 vd->vdev_stat.vs_space = 0;
2960 vd->vdev_stat.vs_dspace = 0;
2961 vd->vdev_stat.vs_alloc = 0;
2962 mutex_exit(&vd->vdev_stat_lock);
2966 vdev_scan_stat_init(vdev_t *vd)
2968 vdev_stat_t *vs = &vd->vdev_stat;
2970 for (int c = 0; c < vd->vdev_children; c++)
2971 vdev_scan_stat_init(vd->vdev_child[c]);
2973 mutex_enter(&vd->vdev_stat_lock);
2974 vs->vs_scan_processed = 0;
2975 mutex_exit(&vd->vdev_stat_lock);
2979 vdev_stat_update(zio_t *zio, uint64_t psize)
2981 spa_t *spa = zio->io_spa;
2982 vdev_t *rvd = spa->spa_root_vdev;
2983 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2985 uint64_t txg = zio->io_txg;
2986 vdev_stat_t *vs = &vd->vdev_stat;
2987 zio_type_t type = zio->io_type;
2988 int flags = zio->io_flags;
2991 * If this i/o is a gang leader, it didn't do any actual work.
2993 if (zio->io_gang_tree)
2996 if (zio->io_error == 0) {
2998 * If this is a root i/o, don't count it -- we've already
2999 * counted the top-level vdevs, and vdev_get_stats() will
3000 * aggregate them when asked. This reduces contention on
3001 * the root vdev_stat_lock and implicitly handles blocks
3002 * that compress away to holes, for which there is no i/o.
3003 * (Holes never create vdev children, so all the counters
3004 * remain zero, which is what we want.)
3006 * Note: this only applies to successful i/o (io_error == 0)
3007 * because unlike i/o counts, errors are not additive.
3008 * When reading a ditto block, for example, failure of
3009 * one top-level vdev does not imply a root-level error.
3014 ASSERT(vd == zio->io_vd);
3016 if (flags & ZIO_FLAG_IO_BYPASS)
3019 mutex_enter(&vd->vdev_stat_lock);
3021 if (flags & ZIO_FLAG_IO_REPAIR) {
3022 if (flags & ZIO_FLAG_SCAN_THREAD) {
3023 dsl_scan_phys_t *scn_phys =
3024 &spa->spa_dsl_pool->dp_scan->scn_phys;
3025 uint64_t *processed = &scn_phys->scn_processed;
3028 if (vd->vdev_ops->vdev_op_leaf)
3029 atomic_add_64(processed, psize);
3030 vs->vs_scan_processed += psize;
3033 if (flags & ZIO_FLAG_SELF_HEAL)
3034 vs->vs_self_healed += psize;
3038 vs->vs_bytes[type] += psize;
3040 mutex_exit(&vd->vdev_stat_lock);
3044 if (flags & ZIO_FLAG_SPECULATIVE)
3048 * If this is an I/O error that is going to be retried, then ignore the
3049 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3050 * hard errors, when in reality they can happen for any number of
3051 * innocuous reasons (bus resets, MPxIO link failure, etc).
3053 if (zio->io_error == EIO &&
3054 !(zio->io_flags & ZIO_FLAG_IO_RETRY))
3058 * Intent logs writes won't propagate their error to the root
3059 * I/O so don't mark these types of failures as pool-level
3062 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
3065 mutex_enter(&vd->vdev_stat_lock);
3066 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
3067 if (zio->io_error == ECKSUM)
3068 vs->vs_checksum_errors++;
3070 vs->vs_read_errors++;
3072 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
3073 vs->vs_write_errors++;
3074 mutex_exit(&vd->vdev_stat_lock);
3076 if (type == ZIO_TYPE_WRITE && txg != 0 &&
3077 (!(flags & ZIO_FLAG_IO_REPAIR) ||
3078 (flags & ZIO_FLAG_SCAN_THREAD) ||
3079 spa->spa_claiming)) {
3081 * This is either a normal write (not a repair), or it's
3082 * a repair induced by the scrub thread, or it's a repair
3083 * made by zil_claim() during spa_load() in the first txg.
3084 * In the normal case, we commit the DTL change in the same
3085 * txg as the block was born. In the scrub-induced repair
3086 * case, we know that scrubs run in first-pass syncing context,
3087 * so we commit the DTL change in spa_syncing_txg(spa).
3088 * In the zil_claim() case, we commit in spa_first_txg(spa).
3090 * We currently do not make DTL entries for failed spontaneous
3091 * self-healing writes triggered by normal (non-scrubbing)
3092 * reads, because we have no transactional context in which to
3093 * do so -- and it's not clear that it'd be desirable anyway.
3095 if (vd->vdev_ops->vdev_op_leaf) {
3096 uint64_t commit_txg = txg;
3097 if (flags & ZIO_FLAG_SCAN_THREAD) {
3098 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3099 ASSERT(spa_sync_pass(spa) == 1);
3100 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
3101 commit_txg = spa_syncing_txg(spa);
3102 } else if (spa->spa_claiming) {
3103 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
3104 commit_txg = spa_first_txg(spa);
3106 ASSERT(commit_txg >= spa_syncing_txg(spa));
3107 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
3109 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
3110 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
3111 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
3114 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
3119 * Update the in-core space usage stats for this vdev, its metaslab class,
3120 * and the root vdev.
3123 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
3124 int64_t space_delta)
3126 int64_t dspace_delta = space_delta;
3127 spa_t *spa = vd->vdev_spa;
3128 vdev_t *rvd = spa->spa_root_vdev;
3129 metaslab_group_t *mg = vd->vdev_mg;
3130 metaslab_class_t *mc = mg ? mg->mg_class : NULL;
3132 ASSERT(vd == vd->vdev_top);
3135 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
3136 * factor. We must calculate this here and not at the root vdev
3137 * because the root vdev's psize-to-asize is simply the max of its
3138 * childrens', thus not accurate enough for us.
3140 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
3141 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
3142 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
3143 vd->vdev_deflate_ratio;
3145 mutex_enter(&vd->vdev_stat_lock);
3146 vd->vdev_stat.vs_alloc += alloc_delta;
3147 vd->vdev_stat.vs_space += space_delta;
3148 vd->vdev_stat.vs_dspace += dspace_delta;
3149 mutex_exit(&vd->vdev_stat_lock);
3151 if (mc == spa_normal_class(spa)) {
3152 mutex_enter(&rvd->vdev_stat_lock);
3153 rvd->vdev_stat.vs_alloc += alloc_delta;
3154 rvd->vdev_stat.vs_space += space_delta;
3155 rvd->vdev_stat.vs_dspace += dspace_delta;
3156 mutex_exit(&rvd->vdev_stat_lock);
3160 ASSERT(rvd == vd->vdev_parent);
3161 ASSERT(vd->vdev_ms_count != 0);
3163 metaslab_class_space_update(mc,
3164 alloc_delta, defer_delta, space_delta, dspace_delta);
3169 * Mark a top-level vdev's config as dirty, placing it on the dirty list
3170 * so that it will be written out next time the vdev configuration is synced.
3171 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
3174 vdev_config_dirty(vdev_t *vd)
3176 spa_t *spa = vd->vdev_spa;
3177 vdev_t *rvd = spa->spa_root_vdev;
3180 ASSERT(spa_writeable(spa));
3183 * If this is an aux vdev (as with l2cache and spare devices), then we
3184 * update the vdev config manually and set the sync flag.
3186 if (vd->vdev_aux != NULL) {
3187 spa_aux_vdev_t *sav = vd->vdev_aux;
3191 for (c = 0; c < sav->sav_count; c++) {
3192 if (sav->sav_vdevs[c] == vd)
3196 if (c == sav->sav_count) {
3198 * We're being removed. There's nothing more to do.
3200 ASSERT(sav->sav_sync == B_TRUE);
3204 sav->sav_sync = B_TRUE;
3206 if (nvlist_lookup_nvlist_array(sav->sav_config,
3207 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
3208 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
3209 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
3215 * Setting the nvlist in the middle if the array is a little
3216 * sketchy, but it will work.
3218 nvlist_free(aux[c]);
3219 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
3225 * The dirty list is protected by the SCL_CONFIG lock. The caller
3226 * must either hold SCL_CONFIG as writer, or must be the sync thread
3227 * (which holds SCL_CONFIG as reader). There's only one sync thread,
3228 * so this is sufficient to ensure mutual exclusion.
3230 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3231 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3232 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3235 for (c = 0; c < rvd->vdev_children; c++)
3236 vdev_config_dirty(rvd->vdev_child[c]);
3238 ASSERT(vd == vd->vdev_top);
3240 if (!list_link_active(&vd->vdev_config_dirty_node) &&
3242 list_insert_head(&spa->spa_config_dirty_list, vd);
3247 vdev_config_clean(vdev_t *vd)
3249 spa_t *spa = vd->vdev_spa;
3251 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
3252 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3253 spa_config_held(spa, SCL_CONFIG, RW_READER)));
3255 ASSERT(list_link_active(&vd->vdev_config_dirty_node));
3256 list_remove(&spa->spa_config_dirty_list, vd);
3260 * Mark a top-level vdev's state as dirty, so that the next pass of
3261 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
3262 * the state changes from larger config changes because they require
3263 * much less locking, and are often needed for administrative actions.
3266 vdev_state_dirty(vdev_t *vd)
3268 spa_t *spa = vd->vdev_spa;
3270 ASSERT(spa_writeable(spa));
3271 ASSERT(vd == vd->vdev_top);
3274 * The state list is protected by the SCL_STATE lock. The caller
3275 * must either hold SCL_STATE as writer, or must be the sync thread
3276 * (which holds SCL_STATE as reader). There's only one sync thread,
3277 * so this is sufficient to ensure mutual exclusion.
3279 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3280 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3281 spa_config_held(spa, SCL_STATE, RW_READER)));
3283 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
3284 list_insert_head(&spa->spa_state_dirty_list, vd);
3288 vdev_state_clean(vdev_t *vd)
3290 spa_t *spa = vd->vdev_spa;
3292 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
3293 (dsl_pool_sync_context(spa_get_dsl(spa)) &&
3294 spa_config_held(spa, SCL_STATE, RW_READER)));
3296 ASSERT(list_link_active(&vd->vdev_state_dirty_node));
3297 list_remove(&spa->spa_state_dirty_list, vd);
3301 * Propagate vdev state up from children to parent.
3304 vdev_propagate_state(vdev_t *vd)
3306 spa_t *spa = vd->vdev_spa;
3307 vdev_t *rvd = spa->spa_root_vdev;
3308 int degraded = 0, faulted = 0;
3312 if (vd->vdev_children > 0) {
3313 for (int c = 0; c < vd->vdev_children; c++) {
3314 child = vd->vdev_child[c];
3317 * Don't factor holes into the decision.
3319 if (child->vdev_ishole)
3322 if (!vdev_readable(child) ||
3323 (!vdev_writeable(child) && spa_writeable(spa))) {
3325 * Root special: if there is a top-level log
3326 * device, treat the root vdev as if it were
3329 if (child->vdev_islog && vd == rvd)
3333 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
3337 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
3341 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
3344 * Root special: if there is a top-level vdev that cannot be
3345 * opened due to corrupted metadata, then propagate the root
3346 * vdev's aux state as 'corrupt' rather than 'insufficient
3349 if (corrupted && vd == rvd &&
3350 rvd->vdev_state == VDEV_STATE_CANT_OPEN)
3351 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
3352 VDEV_AUX_CORRUPT_DATA);
3355 if (vd->vdev_parent)
3356 vdev_propagate_state(vd->vdev_parent);
3360 * Set a vdev's state. If this is during an open, we don't update the parent
3361 * state, because we're in the process of opening children depth-first.
3362 * Otherwise, we propagate the change to the parent.
3364 * If this routine places a device in a faulted state, an appropriate ereport is
3368 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
3370 uint64_t save_state;
3371 spa_t *spa = vd->vdev_spa;
3373 if (state == vd->vdev_state) {
3374 vd->vdev_stat.vs_aux = aux;
3378 save_state = vd->vdev_state;
3380 vd->vdev_state = state;
3381 vd->vdev_stat.vs_aux = aux;
3384 * If we are setting the vdev state to anything but an open state, then
3385 * always close the underlying device unless the device has requested
3386 * a delayed close (i.e. we're about to remove or fault the device).
3387 * Otherwise, we keep accessible but invalid devices open forever.
3388 * We don't call vdev_close() itself, because that implies some extra
3389 * checks (offline, etc) that we don't want here. This is limited to
3390 * leaf devices, because otherwise closing the device will affect other
3393 if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
3394 vd->vdev_ops->vdev_op_leaf)
3395 vd->vdev_ops->vdev_op_close(vd);
3397 if (vd->vdev_removed &&
3398 state == VDEV_STATE_CANT_OPEN &&
3399 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
3401 * If the previous state is set to VDEV_STATE_REMOVED, then this
3402 * device was previously marked removed and someone attempted to
3403 * reopen it. If this failed due to a nonexistent device, then
3404 * keep the device in the REMOVED state. We also let this be if
3405 * it is one of our special test online cases, which is only
3406 * attempting to online the device and shouldn't generate an FMA
3409 vd->vdev_state = VDEV_STATE_REMOVED;
3410 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
3411 } else if (state == VDEV_STATE_REMOVED) {
3412 vd->vdev_removed = B_TRUE;
3413 } else if (state == VDEV_STATE_CANT_OPEN) {
3415 * If we fail to open a vdev during an import or recovery, we
3416 * mark it as "not available", which signifies that it was
3417 * never there to begin with. Failure to open such a device
3418 * is not considered an error.
3420 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
3421 spa_load_state(spa) == SPA_LOAD_RECOVER) &&
3422 vd->vdev_ops->vdev_op_leaf)
3423 vd->vdev_not_present = 1;
3426 * Post the appropriate ereport. If the 'prevstate' field is
3427 * set to something other than VDEV_STATE_UNKNOWN, it indicates
3428 * that this is part of a vdev_reopen(). In this case, we don't
3429 * want to post the ereport if the device was already in the
3430 * CANT_OPEN state beforehand.
3432 * If the 'checkremove' flag is set, then this is an attempt to
3433 * online the device in response to an insertion event. If we
3434 * hit this case, then we have detected an insertion event for a
3435 * faulted or offline device that wasn't in the removed state.
3436 * In this scenario, we don't post an ereport because we are
3437 * about to replace the device, or attempt an online with
3438 * vdev_forcefault, which will generate the fault for us.
3440 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3441 !vd->vdev_not_present && !vd->vdev_checkremove &&
3442 vd != spa->spa_root_vdev) {
3446 case VDEV_AUX_OPEN_FAILED:
3447 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3449 case VDEV_AUX_CORRUPT_DATA:
3450 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3452 case VDEV_AUX_NO_REPLICAS:
3453 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3455 case VDEV_AUX_BAD_GUID_SUM:
3456 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3458 case VDEV_AUX_TOO_SMALL:
3459 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3461 case VDEV_AUX_BAD_LABEL:
3462 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3465 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3468 zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3471 /* Erase any notion of persistent removed state */
3472 vd->vdev_removed = B_FALSE;
3474 vd->vdev_removed = B_FALSE;
3478 * Notify the fmd of the state change. Be verbose and post
3479 * notifications even for stuff that's not important; the fmd agent can
3480 * sort it out. Don't emit state change events for non-leaf vdevs since
3481 * they can't change state on their own. The FMD can check their state
3482 * if it wants to when it sees that a leaf vdev had a state change.
3484 if (vd->vdev_ops->vdev_op_leaf)
3485 zfs_post_state_change(spa, vd);
3487 if (!isopen && vd->vdev_parent)
3488 vdev_propagate_state(vd->vdev_parent);
3492 * Check the vdev configuration to ensure that it's capable of supporting
3493 * a root pool. We do not support partial configuration.
3494 * In addition, only a single top-level vdev is allowed.
3496 * FreeBSD does not have above limitations.
3499 vdev_is_bootable(vdev_t *vd)
3502 if (!vd->vdev_ops->vdev_op_leaf) {
3503 char *vdev_type = vd->vdev_ops->vdev_op_type;
3505 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3506 vd->vdev_children > 1) {
3508 } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3513 for (int c = 0; c < vd->vdev_children; c++) {
3514 if (!vdev_is_bootable(vd->vdev_child[c]))
3517 #endif /* illumos */
3522 * Load the state from the original vdev tree (ovd) which
3523 * we've retrieved from the MOS config object. If the original
3524 * vdev was offline or faulted then we transfer that state to the
3525 * device in the current vdev tree (nvd).
3528 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3530 spa_t *spa = nvd->vdev_spa;
3532 ASSERT(nvd->vdev_top->vdev_islog);
3533 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3534 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3536 for (int c = 0; c < nvd->vdev_children; c++)
3537 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3539 if (nvd->vdev_ops->vdev_op_leaf) {
3541 * Restore the persistent vdev state
3543 nvd->vdev_offline = ovd->vdev_offline;
3544 nvd->vdev_faulted = ovd->vdev_faulted;
3545 nvd->vdev_degraded = ovd->vdev_degraded;
3546 nvd->vdev_removed = ovd->vdev_removed;
3551 * Determine if a log device has valid content. If the vdev was
3552 * removed or faulted in the MOS config then we know that
3553 * the content on the log device has already been written to the pool.
3556 vdev_log_state_valid(vdev_t *vd)
3558 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3562 for (int c = 0; c < vd->vdev_children; c++)
3563 if (vdev_log_state_valid(vd->vdev_child[c]))
3570 * Expand a vdev if possible.
3573 vdev_expand(vdev_t *vd, uint64_t txg)
3575 ASSERT(vd->vdev_top == vd);
3576 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3578 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3579 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3580 vdev_config_dirty(vd);
3588 vdev_split(vdev_t *vd)
3590 vdev_t *cvd, *pvd = vd->vdev_parent;
3592 vdev_remove_child(pvd, vd);
3593 vdev_compact_children(pvd);
3595 cvd = pvd->vdev_child[0];
3596 if (pvd->vdev_children == 1) {
3597 vdev_remove_parent(cvd);
3598 cvd->vdev_splitting = B_TRUE;
3600 vdev_propagate_state(cvd);
3604 vdev_deadman(vdev_t *vd)
3606 for (int c = 0; c < vd->vdev_children; c++) {
3607 vdev_t *cvd = vd->vdev_child[c];
3612 if (vd->vdev_ops->vdev_op_leaf) {
3613 vdev_queue_t *vq = &vd->vdev_queue;
3615 mutex_enter(&vq->vq_lock);
3616 if (avl_numnodes(&vq->vq_active_tree) > 0) {
3617 spa_t *spa = vd->vdev_spa;
3622 * Look at the head of all the pending queues,
3623 * if any I/O has been outstanding for longer than
3624 * the spa_deadman_synctime we panic the system.
3626 fio = avl_first(&vq->vq_active_tree);
3627 delta = gethrtime() - fio->io_timestamp;
3628 if (delta > spa_deadman_synctime(spa)) {
3629 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, "
3630 "delta %lluns, last io %lluns",
3631 fio->io_timestamp, delta,
3632 vq->vq_io_complete_ts);
3633 fm_panic("I/O to pool '%s' appears to be "
3634 "hung on vdev guid %llu at '%s'.",
3636 (long long unsigned int) vd->vdev_guid,
3640 mutex_exit(&vq->vq_lock);